Towards Supervised Robot-assisted Physical Therapy after Hand Fractures

This paper reports on robotic support for physical therapists during rehabilitation after hand fractures. The robotic assistance is based on a humanoid robot Pepper, which is configured to demonstrate and eventually supervise 18 standard rehabilitation exercises. Pepper first explains to a patient how an exercise should be performed, then repeats the instructions and counts how many times an exercise has been correctly repeated. At the same time, he converses with the patient to increase motivation to continue exercising. A neural network based approach has been proposed to automatically detect the patient’s performance. It was found that by using the collected datasets (images) to interpret American and Arabic sign languages, it is possible to obtain a reliable feedback on the exercise performance, allowing a decision on the number of successful repetitions for each exercise. Finally, some challenges encountered are highlighted and their possible solutions are discussed.